1
0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 07:43:03 +04:00

check FP16 build condition correctly

* use __GNUC_MINOR__ in correct place to check the version of GCC
  * check processor support of FP16 at run time
  * check compiler support of FP16 and pass correct compiler option
  * rely on ENABLE_AVX on gcc since AVX is generated when mf16c is passed
  * guard correctly using ifdef in case of various configuration
  * use v_float16x4 correctly by including the right header file
This commit is contained in:
Tomoaki Teshima
2016-09-21 15:42:54 +09:00
committed by Tomoaki Teshima
parent d5c202e46b
commit c7cb116dc0
4 changed files with 68 additions and 46 deletions
+32 -24
View File
@@ -711,48 +711,56 @@ template<typename R> struct TheTest
return *this;
}
#if CV_FP16
TheTest & test_loadstore_fp16()
{
#if CV_FP16
AlignedData<R> data;
AlignedData<R> out;
// check if addresses are aligned and unaligned respectively
EXPECT_EQ((size_t)0, (size_t)&data.a.d % 16);
EXPECT_NE((size_t)0, (size_t)&data.u.d % 16);
EXPECT_EQ((size_t)0, (size_t)&out.a.d % 16);
EXPECT_NE((size_t)0, (size_t)&out.u.d % 16);
if(checkHardwareSupport(CV_CPU_FP16))
{
// check if addresses are aligned and unaligned respectively
EXPECT_EQ((size_t)0, (size_t)&data.a.d % 16);
EXPECT_NE((size_t)0, (size_t)&data.u.d % 16);
EXPECT_EQ((size_t)0, (size_t)&out.a.d % 16);
EXPECT_NE((size_t)0, (size_t)&out.u.d % 16);
// check some initialization methods
R r1 = data.u;
R r2 = v_load_f16(data.a.d);
R r3(r2);
EXPECT_EQ(data.u[0], r1.get0());
EXPECT_EQ(data.a[0], r2.get0());
EXPECT_EQ(data.a[0], r3.get0());
// check some initialization methods
R r1 = data.u;
R r2 = v_load_f16(data.a.d);
R r3(r2);
EXPECT_EQ(data.u[0], r1.get0());
EXPECT_EQ(data.a[0], r2.get0());
EXPECT_EQ(data.a[0], r3.get0());
// check some store methods
out.a.clear();
v_store_f16(out.a.d, r1);
EXPECT_EQ(data.a, out.a);
// check some store methods
out.a.clear();
v_store_f16(out.a.d, r1);
EXPECT_EQ(data.a, out.a);
}
return *this;
#endif
}
TheTest & test_float_cvt_fp16()
{
#if CV_FP16
AlignedData<v_float32x4> data;
// check conversion
v_float32x4 r1 = v_load(data.a.d);
v_float16x4 r2 = v_cvt_f16(r1);
v_float32x4 r3 = v_cvt_f32(r2);
EXPECT_EQ(0x3c00, r2.get0());
EXPECT_EQ(r3.get0(), r1.get0());
if(checkHardwareSupport(CV_CPU_FP16))
{
// check conversion
v_float32x4 r1 = v_load(data.a.d);
v_float16x4 r2 = v_cvt_f16(r1);
v_float32x4 r3 = v_cvt_f32(r2);
EXPECT_EQ(0x3c00, r2.get0());
EXPECT_EQ(r3.get0(), r1.get0());
}
return *this;
}
#endif
}
};